Behavioral Ecology doi:10.1093/beheco/arm151 Advance Access publication 11 January 2008 Can experienced birds select for Mu¨llerian mimicry?

Eira Ihalainen, Leena Lindstro¨m, Johanna Mappes, and Sari Puolakkainen Department of Biological and Environmental Science, PO Box 35, FI-40014, University of Jyva¨skyla¨, Finland Downloaded from https://academic.oup.com/beheco/article/19/2/362/214147 by guest on 27 September 2021

Field experiments have shown that avian predators in the wild can select for similarity of warning signals in aposematic prey (Mu¨llerian mimicry) because a common signal is better protected than a signal that is novel and rare. The original theory of Mu¨llerian mimicry assumes that the mechanism promoting mimicry is predator learning; by sharing a signal, the comimic species share the mortality that is due to sampling by inexperienced predators. Predation events have not been observed in the wild, and learning experiments with naive bird predators in a laboratory have not unambiguously shown a benefit of a uniform signal compared with different signals. As predators in the field experiments are likely to be more experienced compared with previous laboratory experiments, we studied selection by experienced predators on a novel imperfect mimic. We trained great tits Parus major to avoid artificial aposematic models and subsequently introduced perfect and imperfect mimics at different frequencies. Birds with prior experience on the models selected against the imperfect mimics that were at a disadvantage also in a memory test conducted a week after their introduction. Selection against the imperfect mimics was antiapostatic. However, the imperfect mimics also benefited from some signal generalization to the models and possibly gained protection because the birds were familiar with the alternative cryptic prey that was also present. Our results suggest that experienced predators might be more important to the evolution of mimicry than the learning-based theory assumes. Key words: antiapostatic selection, aposematism, generalization, learning, memory. [Behav Ecol 19:362–368 (2008)]

redators can learn to recognize unprofitable prey and re- eficial to all individuals using the signal. In his original formu- Pject them by sight. Aposematic species, which are unprofit- lation, Mu¨ller (1879) used a numerical example of 2 species of able due to, for example, toxicity, have typically conspicuous different population sizes to show how the benefit of mimicry is appearance (Poulton 1890). Conspicuous coloration can act relatively greater to a rarer species if predators sample a fixed as a warning signal that identifies the prey as unprofitable number of prey individuals from across 2 species. (Wallace 1867; Sherratt and Beatty 2003; Ham et al. 2006; Despite examples that show how signal similarity must be see also Jansson and Enquist 2003) and facilitates the avoid- beneficial to aposematic prey in nature (Benson 1972; Mallet ance learning of predators (Gittleman and Harvey 1980; and Barton 1989; Dumbacher and Fleischer 2001; Kapan Gittleman et al. 1980; Sille´n-Tullberg 1985; Roper and Wistow 2001; Symula et al. 2001), the mechanism promoting mimicry 1986; Alatalo and Mappes 1996). Aposematic prey are typically remains unclear. The learning phase of predators has been expected to suffer highest mortality when rare, and such anti- studied with laboratory experiments. Studies where great tits apostatic selection is considered a barrier for the evolution of (Parus major) were used as inexperienced predators of artifi- aposematic defense strategy (Lindstro¨m et al. 2001b; see also cial prey have not shown constant benefits of a shared warning Endler 1988 and Ruxton et al. 2004 for a review of the ‘‘rarity signal versus different warning signals (Rowe et al. 2004; Ham problem’’). et al. 2006; Lindstro¨m et al. 2006; Ihalainen et al. 2007; but see Mu¨llerian comimics are aposematic species that possess Rowland et al. 2007 and also Beatty et al. 2004). Instead, in a similar warning signal; the best-known examples are insects field studies where live aposematic prey was used, transferred with similar color patterns (see, e.g., Mu¨ller 1879; Plowright butterflies with a wing pattern that matched that of the locally and Owen 1980; see also Ruxton et al. 2004). In his original abundant morph survived better than individuals with a differ- theory of Mu¨llerian mimicry, Mu¨ller (1879) proposed that ent wing pattern (Benson 1972; Mallet and Barton 1989; a similar warning signal is beneficial for the comimics because Kapan 2001), and this difference in survival was likely due species that mimic each other share the mortality that is due to predation by birds. In the wild, even simple predator com- to ‘‘predator education.’’ This theory is in line with the obser- munities can be heterogeneous in the level of experience the vation of antiapostatic selection on aposematic prey (Green- individuals have (see Endler and Mappes 2004). Although wood et al. 1989; Lindstro¨m et al. 2001b). If naive predators predation was never directly observed in the experiments have to sample a certain number of individuals with a given where butterflies were transferred between locations, it is pos- signal to learn to reject prey with that particular signal by sible that the butterflies were attacked not solely by naive sight, comimics that use the same warning signal gain by re- predators that were learning to avoid aposematic prey but ducing their per capita risk of getting sampled; the same by birds that were already familiar with the local butterfly amount of individuals that is needed to educate the predators morphs and did not generalize their learned avoidances to is divided between the mimetic species. A common appear- the transferred morphs. Experienced jacamars, which were ance is thus mutually (although not necessarily equally) ben- potential predators of the transferred butterflies (see Mallet and Barton 1989), have indeed been shown to attack novel Address correspondence to E. Ihalainen. E-mail: eiraihal@jyu.fi. aposematic prey (Langham 2006). Kapan (2001) also found Received 20 June 2007; revised 19 October 2007; accepted 19 some evidence that selection against the introduced morphs November 2007. was relaxed when they were released in a higher density, which also emphasizes the principle that aposematic species and The Author 2008. Published by Oxford University Press on behalf of the International Society for Behavioral Ecology. All rights reserved. For permissions, please e-mail: [email protected] Ihalainen et al. • Predator experience and Mu¨llerian mimicry 363 Downloaded from https://academic.oup.com/beheco/article/19/2/362/214147 by guest on 27 September 2021

Figure 1 The experimental setup, number of birds, and the types and numbers of prey items presented in the learning trial, mimic trial, and memory test. Edible cryptic prey (cr) was marked with the cross signal. The star and square signals alternated as models (mo), perfect mimics (pe), and imperfect mimics (im). comimics have ‘‘strength in numbers’’ (see also Rowland et al. leased back into the wild at the location of their capture. They 2007). were housed individually in illuminated plywood cages, sized The field experiments differ from the learning experiments 64 3 46 3 77 cm with a daily light period of 11 h. Sunflower in their starting point: in field experiments, the predator com- seeds, peanuts, tallow, and fresh water were available ad libi- munity is presented with a new morph (in addition to natu- tum, and occasionally, the birds were offered mealworms (lar- rally occurring species that they have previous experience of), vae of Tenebrio molitor). Prior to the experiment, the birds were whereas learning experiments present naive predators with food deprived (ca. 2 h) to ensure motivation to search for the different morphs simultaneously. Thus, the predators have artificial prey but water was always available. had different level of experience in these studies; the field experiments focus more on generalization, and the laboratory Artificial prey and experimental aviaries experiments focus on the learning process. Here, we present data from a laboratory experiment where The prey items were 8 3 8 mm paper shells with small pieces bird predators had previous experience on aposematic prey. (ca. 0.1 g in weight) of almond glued inside. We created 2 Great tits were first trained to forage on artificial prey and to aposematic prey types: models/perfect mimics and imperfect discriminate between cryptic, edible prey, and highly unpalat- mimics. These prey items were made unpalatable by soaking able aposematic prey (models). Subsequently, we presented the almond for an hour in a solution of 30 mL of water and the birds with a perfect mimic (identical to the model) and 2 g of chloroquine phosphate (models and perfect mimics) or an imperfect mimic that was not only visually different to the 0.25 g of chloroquine phosphate (imperfect mimics). These model (and thus unfamiliar to the birds) but also less aversive concentrations translate to a difference in taste to the birds in taste. We presented the mimics at different frequencies. A (Lindstro¨m et al. 2006; Ihalainen et al. 2007). The birds typ- week later, we tested the level of avoidance the birds had for ically react to the bitter taste of chloroquine phosphate by the signaling prey in a ‘‘memory test.’’ The results show that head shaking, beak wiping, and drinking. The difference in when studying predation on mimetic prey the context of pre- unpalatability level was included because it is assumed that sentation is important and that experienced predators could a species can be a ‘‘mimic’’ that evolves to resemble a ‘‘model’’ readily select for accurate Mu¨llerian mimicry. in a Mu¨llerian mimicry complex if it is less defended (see Mallet 1999). It has been previously shown that compared METHODS with highly unpalatable prey, learning rate is slower (and thus mortality is higher) for moderately unpalatable prey in this The experiments were conducted at the Konnevesi Research system (Lindstro¨m et al. 2006; Ihalainen et al. 2007). There- Station in Central Finland from September to December fore, we expected the selection pressure on the imperfect 2001. Capturing and using the birds in the study was licensed mimic to be a conservative upper limit estimate. The cryptic by the Central Finland Regional Environment Center (permis- edible prey contained a piece of untreated almond. sion number LS-46/01, 0901L0448/254) and the Experimen- A black-and-white signal was printed on both sides of all tal Animal Committee of the University of Jyva¨skyla¨ types of prey items. The cryptic pattern was a cross symbol (permission number 19/5.6.2001). that was also printed on the background of the aviary floor (see below and Figure 1). The 2 different warning signals were Birds a square and an asymmetrical star (see Figure 1). These sig- Great tits were caught from feeding sites with traps and ringed nals were equally visible, and great tits did not show prefer- for identification. After the experiments, the birds were re- ence for either of them when palatable (Ihalainen 2006). 364 Behavioral Ecology

The aviary background consisted of white paper with black trial did not differ between the frequency treatments (ANOVA crosses printed on it. The paper was covered with adhesive F2,58 ¼ 0.067, P ¼ 0.935), which ensured that initial differ- plastic, and ‘‘fake prey items’’ (printed crosses cut out of white ences in avoidance or the amount of negative experience cardboard) were glued onto it to make the background 3 di- would not confound the possible effect of mimic frequencies mensional. The background formed a grid of 22 columns and on prey mortalities (for further analysis of the whole learning 15 rows, that is, 330 cells. Wooden boards were placed be- trial, see Results). The birds were again allowed to kill 30 prey tween the rows to ease prey handling and movement of the items (trials lasted on average ca. 53 min), and the total num- birds. The learning and mimic trials (see below) were run in ber of each prey type killed was recorded. an aviary with a floor area of 57 m2 (the black-and-white back- The experiment was therefore a 2 3 3 design (signal of the ground covered 41 m2 of this). There were 8 perches at the model 3 frequencies of the mimics), where there were 12 height of 0.5 m for prey handling. For the memory test and birds in the 95/5 frequency treatments and 10 birds in the training of the birds (see below), we used 2 smaller aviaries other 4 treatments, so 64 birds altogether (Figure 1).

where the paper background formed a grid of 8 rows and Downloaded from https://academic.oup.com/beheco/article/19/2/362/214147 by guest on 27 September 2021 10 columns and covered an area of 10 m2. These aviaries had Memory test 2 perches. All aviaries were illuminated with energy-saving light The memory test was carried out in the small aviaries (see bulbs, and fresh water was available to the birds. Methods) a week after the mimic trial. We included 43 of The birds were observed through a 1-way window during the 44 birds from the 95/5 and 50/50 frequency treatments the trials. We used the coordinates of the grid to identify only (1 bird died before this test). We put 30 cryptic prey the prey type that was attacked by the bird as individual cells items, 15 squares and 15 stars, on the background grid of 80 of the grid were occupied by 1 prey item only (see below). cells in a random order. The birds were allowed to kill 15 prey items (completing the test took on average ca. 28 min), and Pretraining and learning trial the total number of each prey type killed was recorded. All the prey items were palatable so that the birds could not All birds (n ¼ 64) were first trained stepwise to handle the strengthen their avoidance of the mimetic prey but could have artificial prey in their home cages. Blank white paper shells relearned to accept them as edible prey. We did not detect were used for the training prey. The birds were also familiar- such a trend when comparing the number of signaling prey ized with the aviaries and trained to forage from the black- items the birds took within the first and last 5 prey items they and-white background to ensure they would start foraging in were allowed to kill during the memory test: a total of 13 birds the actual trial and could utilize the perches. To familiarize increased their attacks on the signaling prey, and 13 birds the birds with the large aviary, several individuals at a time decreased their attacks as the test progressed. The remaining were allowed to feed and overnight there. The background 17 birds took equal numbers of signaling prey in the begin- grid was replaced with transparent plastic, and sunflower ning and in the end of the test. Thus, we are not measuring seeds and peanuts were spread on it. The small aviaries were the rate of memory extinction, but the attacks on models/ used for background training during which the birds had to perfect mimics and imperfect mimics in this simple test reflect find and eat 12 plain white and 3 cryptic prey items (crosses). the ‘‘level of learning’’ or the willingness of the birds to sam- These were included to give the birds some experience with ple the signaling prey items after the preceding trials. the cryptic prey before first encounter with the defended prey (for details of training, see Ihalainen et al. 2007). Data analysis In the learning trial, 100 highly unpalatable model prey items and 100 cryptic prey items were randomly distributed For the analyses, we calculated a relative predation risk for the on the background grid. For half of the birds the square was models (learning trial) and mimics (mimic trial and memory the signal of the model, and for the other half it was the star. test) by dividing the number of killed prey items of each type by To gain experience on aposematic prey and to learn to avoid the number expected by chance. For example, in the learning the model, the birds were allowed to kill 30 prey items. On trial, when 100 models and 100 cryptic prey items were pre- average, it took 1h18min for the birds to complete a trial. A sented in the setup, the expected number of models killed was prey item was considered to be killed when the bird opened 15 because the birds were allowed to kill 30 prey items. Because the paper shell and tasted or ate the almond inside. We re- the total number of prey presented in each treatment was con- corded the number of each prey type killed. stant, the relative risks give the same results as mortalities (pro- portion of each prey type killed) but on a different scale, where 1 denotes random predation. Random predation does not take Mimic trial into account the visibility difference of the aposematic and cryp- The mimic trial was run on the day after the learning trial. tic prey (i.e., likelihood of detection) but gives a reference point Hundred unpalatable prey items were presented: the unfamil- for comparing the calculated risks of the aposematic prey types. iar mildly unpalatable imperfect mimic (with the signal that For some analyses, we similarly calculated relative predation had not been the model in the learning trial) was introduced risks for the models and mimics in the beginning and/or the alongside the perfect mimics that were identical to the famil- end of a trial, that is, during the first or last 5 prey items killed. iar models. The frequencies of the signaling prey were 95/5 or In those cases, the expected number killed by chance was calcu- 80/20 or 50/50 for the perfect and the imperfect mimics, lated for each bird separately based on what was left in the setup respectively. There were also a 100 cryptic prey items available. for the bird to kill. The data were analyzed with SPSS 11.5 When dividing the birds to the frequency treatments, we took for Windows statistical program. All tests are 2 tailed, and we into account their ‘‘strength of avoidance,’’ that is, the num- used ANOVA by ranks and nonparametric tests where the data ber of models the birds killed in the end of the learning trial did not meet the requirements of parametric tests. (within the last 10 prey items attacked); there was no differ- ence in avoidance between the frequency treatments (analysis RESULTS of variance [ANOVA] F ¼ 0.326, P ¼ 0.723; also no effect of 2,58 Learning trial signal treatment F1,58 ¼ 2.502, P ¼ 0.119 and no interaction between the treatments F2,58 ¼ 0.329, P ¼ 0.721). Further- In the learning trial, when only cryptic and model prey items more, the total number of models killed in the whole learning were offered, the predation risk of the star models was lower Ihalainen et al. • Predator experience and Mu¨llerian mimicry 365 Downloaded from https://academic.oup.com/beheco/article/19/2/362/214147 by guest on 27 September 2021

Figure 2 Figure 3 Mean relative predation risk (6SE) of star and square models in the Mean relative predation risks (6SE) of perfect (black squares) and learning trial. imperfect mimics (gray circles) (6SE) in the mimic trial. The fre- quency treatments are 95 perfect and 5 imperfect mimics presented, 80 perfect and 20 imperfect mimics presented, and 50 perfect and than that of the square models (F1,62 ¼ 5.213, P ¼ 0.026) 50 imperfect mimics presented. Reference line shows random pre- showing that the star was more efficient a warning signal dation (see Data analysis). (Figure 2). There was neither effect of frequency treatment (F2,58 ¼ 0.067, P ¼ 0.935 as the strength of avoidance between the frequency treatments was controlled for, see Methods) nor Z ¼1.452, P ¼ 0.147), which indicates that the unfamiliar 2-way interaction between the signal and frequency treatments morph was at some disadvantage when rare (Figure 3). (F2,58 ¼ 0.342, P ¼ 0.712). The birds learned to avoid the The imperfect mimics seemed to gain protection from models: the mean predation risk for the star models was some generalization to the models. We compared the initial 1.28 (standard error [SE] ¼ 0.09) at the beginning of the trial risk (relative predation risk within the first 5 prey items killed) (during first 5 prey items killed) and decreased to 0.64 (SE ¼ of the imperfect mimics with the initial risk the models had in 0.06) at the end of the trial (during last 5 prey items killed) the beginning of the learning trial (i.e., when they were un- (paired samples t-test t31 ¼ 5.698, P , 0.001). The mean familiar to the birds and likely attacked according to their risk of square models decreased from 1.54 (SE ¼ 0.07) to visibility). We compared star mimics with star models and 0.70 (SE ¼ 0.07) (t31¼ 8.235, P , 0.001). square mimics with square models because the signals differed in mortality in the learning trial; the comparison was thus made between birds in different signal treatments (frequency Mimic trial treatments were combined). The initial risk of the imperfect In the mimic trial, the predation risk of the perfect mimics mimics was lower than that of the models (Mann–Whitney U (that were identical to the models presented in the learning test Z ¼2.221, P ¼ 0.027 for stars and Z ¼3.858, P , 0.001 trial) did not differ between the signal treatments (ANOVA for squares; Figure 4). F1,58 ¼ 2.538, P ¼ 0.117). The frequency treatment did not Despite the milder taste of the imperfect mimics (see affect the risk of perfect mimics (F2,58 ¼ 0.391, P ¼ 0.678). Methods), their overall predation risk during the whole mimic Thus, familiar looking prey did not suffer from increased trial was not higher than the overall risk of the models in the frequency of imperfect mimics and/or their own reduced learning trial: the risk of imperfect star mimics did not differ frequency. There was also no 2-way interaction between the from the risk of star models (Mann–Whitney U test Z ¼ treatments (F2,58 ¼ 0.158, P ¼ 0.855). 1.434, P ¼ 0.152), and the squares survived better as imper- The predation risk of the imperfect mimics was higher than fect mimics than as models (Z ¼2.547, P ¼ 0.011). that of the perfect mimics in all frequency treatments (paired samples t-test, all P , 0.007; Figure 3), which indicates that Memory test familiar aposematic prey had a benefit over unfamiliar prey; educated birds selected against the imperfect mimics. The In the following, the signaling prey are still termed perfect predation risk of the imperfect mimics did not depend on and imperfect mimics according to the treatments in the their frequency (ANOVA by ranks H2,58 ¼ 1.267, P ¼ 0.531) mimic trial even though all prey were palatable in this test. contrary to what would have been expected if selection on the Similarly, frequency treatment also refers to the frequencies in imperfect mimics was antiapostatic. Signal treatment had no the preceding mimic trial because the frequency of both effect on the risk of imperfect mimics (H1,58 ¼ 0.103, P ¼ co-mimics was 15 in this test. 0.748), and there was no 2-way interaction between the fre- The frequency treatment had no subsequent effect on the quency and signal treatments (H2,58 ¼ 3.020, P ¼ 0.221). predation risks of the perfect (ANOVA by ranks H1,39 ¼ 0.866, However, when compared with random predation, only the P ¼ 0.352) or the imperfect mimics (ANOVA F1,39 ¼ 0.530, imperfect mimics with a frequency of 50 were clearly avoided, P ¼ 0.471; Figure 5). This indicates that the birds that had that is, their risk was lower than 1.0 (1-sample t-test t19 ¼ been presented with only 5 imperfect mimics avoided that 5.284, P ¼ 0.001). Imperfect mimics with a frequency of signal in the memory test as much as birds that had been 20 were slightly avoided (t19 ¼1.942, P ¼ 0.067), but those presented with 50 imperfect mimics. The risks of the mimics with a frequency of 5 were not (Wilcoxon signed-ranks test: were not affected by the signal treatment (H1,39 ¼ 1.075, 366 Behavioral Ecology Downloaded from https://academic.oup.com/beheco/article/19/2/362/214147 by guest on 27 September 2021

Figure 5 Relative predation risks (6SE) of perfect (black squares) and im- perfect mimics (gray circles) in the memory test. The 2 original frequency treatments included in the test are 95 perfect and 5 im- perfect mimics presented in the mimic trial and 50 perfect and 50 imperfect mimics presented in the mimic trial. Reference line shows random predation (see Data analysis).

mon pattern (Benson 1972; Mallet and Barton 1989; Kapan 2001). These results support the principle of Mu¨llerian mim- icry (Mu¨ller 1879) that a shared warning signal protects apo- sematic prey better than different warning signals, although they do not emphasize predator learning as the mechanism selecting for mimicry. There are 2 factors that could have caused the difference in predation risk between the perfect and imperfect mimics in this experiment; the imperfect mimics looked different to the familiar signal, and they were less unpalatable than the mod- els and perfect mimics. Consequently, the birds could have Figure 4 been more motivated to sample the milder imperfect mimics Initial predation risks of (a) the star prey as models and as imperfect as they were recognizable (Lindstro¨m et al. 2006; Ihalainen mimics and (b) the square prey as models and as imperfect mimics. et al. 2007; see Pearce 1997, p. 56–59). The imperfect mimics Boxplots show minimum, maximum, and the upper and lower also seemed to have some disadvantage of rarity (predation risk quartiles around the median. The outliers denoted by open circles was less than random only when the imperfect mimics were were included in the analyses. common; Figure 3), which is in line with the observation that novel aposematic prey face antiapostatic selection (Lindstro¨m P ¼ 0.300 for perfect and F1,39 ¼ 0.916, P ¼ 0.344 for imper- et al. 2001b). fect mimics), and there were no 2-way interactions between Antiapostatic selection was possibly attenuated because the the signal and frequency treatments (H1,39 ¼ 0.640, P ¼ 0.424 birds did generalize between the models and the imperfect for the risk of perfect and F1,39 ¼ 0.005, P ¼ 0.942 for the risk mimics to some extent. In the mimic trial, the initial preda- of imperfect mimics). tion risk of the imperfect mimics was lower than the initial risk Overall, perfect mimics survived better than imperfect of models in the learning trial when the models were unfamil- mimics (paired samples t-test t42 ¼2.862, P ¼ 0.007 for all iar to the birds and were likely attacked according to their treatments combined), but both signaling prey types were high visibility (Figure 4). This indicates that the birds may killed less than randomly (Wilcoxon signed-ranks test against have had altogether less interest in sampling the unfamiliar, random predation of 1.0, P , 0.001, for both mimic types; imperfect mimics after their experience with the models. The Figure 5). models and imperfect mimics also resembled each other considerably more than they resembled cryptic prey, and gen- eralization between them may be a result of simple categori- DISCUSSION zation to edible and inedible prey (see Wallace 1867; Sherratt In the present study, birds with prior experience on a warning and Beatty 2003). Additionally, experience on the cryptic prey signal selected against aposematic prey with an unfamiliar alone could have decreased the birds’ willingness to attack the signal; imperfect mimics had a higher predation risk than mimics in the mimic trial (Lindstro¨m et al. 2001a), and con- the familiar-looking perfect mimics (Figures 3 and 5). Similar sequently, strong antiapostatic selection was not observed results have been found by releasing aposematic butterflies when comparing the predation risks of imperfect mimics di- with a locally common or with a locally novel wing pattern rectly between the frequency treatments. in the wild where predators likely had experience of the lo- The memory test where all prey was palatable showed the cally common warning signal: the butterflies with the novel same general pattern that the imperfect mimics had higher wing pattern had higher mortality than ones with the com- predation risk than the perfect mimics (Figure 3). This too Ihalainen et al. • Predator experience and Mu¨llerian mimicry 367 could be either due to the relatively unfamiliar signal of the imperfect mimics can initially have similar problems in spread- imperfect mimics or due to the birds being more willing to ing than aposematic morphs in general (see Mallet and Joron attack prey they expect to be only moderately unpalatable 1999). However, the magnitude of such problems is likely to (i.e., moderately unpalatable prey could have a higher attack be very context dependent: the spread of a morph within asymptote compared with highly unpalatable prey, see, e.g., a species depends not only on its fitness relative to the wild Speed 1993; Speed 1999). These effects cannot be reliably type but also on whether predators generalize the new mimic separated from these data, but some aspects of the data ques- morph to a familiar candidate model, which depends not only tion the importance of moderate unpalatability: the birds had on the degree of visual similarity but also on other aspects. very little actual experience of the mild taste of the imperfect For example, bird predators may generalize more readily mimics. We compared how many models/perfect mimics from negative than positive experience (Ham et al. 2006), (these were combined as they were identical) and imperfect show peak shifts (Gamberale and Tullberg 1996), or general- mimics the birds had tasted or eaten during the learning and ize asymmetrically between 2 different signals (Goodale and

mimic trials. In the beginning of the memory test, an average Sneddon 1977). It has been suggested that predators should Downloaded from https://academic.oup.com/beheco/article/19/2/362/214147 by guest on 27 September 2021 bird from the 95/5 frequency treatments had experience of generalize more broadly between Mu¨llerian comimics that are (had killed) 19.38 (SE ¼ 1.22) models/perfect mimics and all unpalatable than between unpalatable models and their 0.67 (SE ¼ 0.13) imperfect mimics, whereas an average bird edible (Batesian) mimics (Fisher 1958; Huheey 1988). Also, from the 50/50 treatment had experience on 17.25 (SE ¼ the severity of punishment from eating a model affects how 1.30) models/perfect mimics and 4.8 (SE ¼ 0.51) imperfect broadly the predator generalizes to other signals (Goodale mimics. Whether the birds could have learned from 0.67 or and Sneddon 1977; Darst and Cummings 2006). Some species 4.8 imperfect mimics that their unpalatability is only moder- can be better models for Mu¨llerian mimicry complexes than ate is questionable. others due to, for example, abundance or earlier emergence Furthermore, birds from the different frequency treatments (Mallet 1999), and evolving Mu¨llerian mimics with resem- seemed to have the same ‘‘level of learning’’ in the memory blance to such species can be spread more successfully than test (Figure 5): the different numbers of perfect and imper- other morphs (see Mallet and Joron 1999; Beatty et al. 2004). fect mimics that had been presented to them in the mimic From this perspective, it seems plausible that although trial had no effect on their readiness to attack these signaling a new comimic morph that resembles, for example, a more prey in the future (see also Lindstro¨m et al. 2001b). Although common species would benefit from reduced mortality due to frequencies affected the survival of the prey in the mimic trial, predator education, generalization by experienced predators they seemed to have less effect from the birds‘ perspective: the will also have an effect on what kind of signal mutations will birds’ ‘‘actual experience’’ of models/perfect mimics in the survive. Despite the ongoing discussion (e.g., Mallet 1999; different treatments during the learning and mimic trials was Sherratt et al. 2004) and recent data (Langham 2006; Rowland comparable and the absolute numbers of killed imperfect et al. 2007; present results), the relative importance of naive mimics generally very low. Considering how little negative ex- and experienced predators as selective agents in Mu¨llerian perience the birds had with the imperfect mimics, their pre- mimicry remains unsolved because there are no comprehen- dation risk in the memory test is perhaps surprisingly low, but sive data from the wild on how much they kill unfamiliar here too, the availability of palatable alternative (cryptic) prey aposematic prey. It may be that in seasonal environments pre- is likely to be important (see also Kokko et al. 2003; Sherratt dation pressure on aposematic prey increases considerably et al. 2004) in addition to signal generalization. when naive predators start to feed on their own (Ojala During the learning trial, when the birds were trained to 2006), whereas in environments where breeding is not as sea- avoid the models, the models with the star signal had a lower sonal and the ratio of naive and older predators is different, predation risk than those with the square signal (Figure 2; see experienced individuals may play an essential role in selecting also Ihalainen 2006; Lindstro¨m et al. 2006), that is, the 2 for signal similarity in Mu¨llerian mimicry. warning signals differed in efficacy. Despite this initial differ- ence, the weaker signal did not encourage the birds to keep FUNDING attacking the prey because in the mimic trial and in the mem- ory test the difference in predation between the signal treat- Finnish Centre of Excellence in Evolutionary Ecology and ments had disappeared. Additionally, the risk of the imperfect Finnish Centre of Excellence in Evolutionary Research mimics did not depend on their signal when they were intro- funded by the Academy of Finland (Finnish Cultural Founda- duced. This is interesting because generally the same things tion to E.I.). that promote learning should also promote memory (Shettle- worth 1998, p. 246) so that the characteristics of prey that We are grateful to Helina¨ Nisu for keeping the bird laboratory run- enhance learning are also beneficial in the long term (see, ning and the staff of Konnevesi Research Station for facilities and e.g., Gittleman and Harvey 1980; Roper and Redston 1987; helping hands. We thank our great tits for the results and Hannah Roper 1994 on conspicuousness) at least if foraging decisions Rowland for improving the manuscript. Authors after the first are in are seen simply as a result of learning and forgetting. alphabetical order. The present results suggest that experience of the predators makes an important difference in the previous laboratory and REFERENCES field studies and also that they shed light on slightly different aspects of mimicry: experiments that compare learning rates Alatalo RV, Mappes J. 1996. Tracking the evolution of warning signals. of predators facing 1 signal or 2 signals simultaneously are Nature. 382:708–710. perhaps more mechanistic than studies where very small num- Balogh ACV, Leimar O. 2005. Mu¨llerian mimicry: an examination of bers of prey with an unfamiliar signal are under interest. Ex- Fisher’s theory of gradual evolutionary change. Proc R Soc Lond B. perienced birds seem to select for Mu¨llerian mimicry in the 272:2269–2275. Beatty CD, Beirinckx K, Sherratt TN. 2004. The evolution of mu¨llerian sense that a common signal is a better protection against mimicry in multispecies communities. Nature. 431:63–67. them. On the other hand, it is likely that mimicry evolves Benson WW. 1972. Natural selection for Mu¨llerian mimicry in Heli- through imperfect stages (see Turner 1977; Balogh and Lei- conius erato in Costa Rica. Science. 176:936–938. mar 2005). If imperfect mimics never survive, interspecific Darst CR, Cummings ME. 2006. Predator learning favours mimicry of mimicry is unlikely to evolve. Our results suggest that new a less-toxic model in poison frogs. Nature. 440:208–211. 368 Behavioral Ecology

Dumbacher JP, Fleischer RC. 2001. Phylogenetic evidence for colour Mallet JLB. 1999. Causes and consequences of a lack of coevolution in pattern convergence in toxic pitohuis: Mu¨llerian mimicry in birds? Mu¨llerian mimicry. Evol Ecol. 13:777–806. Proc R Soc Lond B. 268:1971–1976. Mallet JLB, Barton NH. 1989. Strong natural selection in a warning- Endler JA. 1988. Frequency-dependent predation, crypsis and apose- color hybrid zone. Evolution. 43:421–431. matic coloration. Philos Trans R Soc Lond B Biol Sci. 319:505–523. Mu¨ller F. 1879. Ituna and Thyridia: a remarkable case of mimicry in Endler JA, Mappes J. 2004. Predator mixes and the conspicuousness butterflies. Trans Entomol Soc Lond. XX–XXIX. of aposematic signals. Am Nat. 163:532–547. Ojala K. 2006. Variation in defence and its fitness consequences in Fisher RA. 1958. The genetical theory of natural selection. 2nd rev. ed. aposematic animals—interactions among diet, parasites and preda- New York: Dover Publication. tors [PhD dissertation]. Jyva¨skyla¨ (Finland): University of Jyva¨skyla¨. Gamberale G, Tullberg BS. 1996. Evidence for peak-shift in predator Pearce JM. 1997. Animal learning and cognition. An introduction, generalization among aposematic prey. Proc R Soc Lond B. 2nd ed. Hove (UK): Psychology Press. 263:1329–1334. Plowright RC, Owen RE. 1980. The evolutionary significance of bum- Gittleman JL, Harvey PH. 1980. Why are distasteful prey not cryptic? ble bee color pattern: a mimetic interpretation. Evolution. 34:622– Nature. 286:149–150. 637. Gittleman JL, Harvey PH, Greenwood PJ. 1980. The evolution of con- Poulton EB. 1890. The colours of animals, their meaning and use, Downloaded from https://academic.oup.com/beheco/article/19/2/362/214147 by guest on 27 September 2021 spicuous coloration: some experiments in bad taste. Anim Behav. especially considered in the case of insects. London: Kegan Paul, 28:897–899. Trench, Tru¨bner, & CO. Ltd. Goodale MA, Sneddon I. 1977. The effect of distastefulness of the Roper TJ. 1994. Conspicuousness of prey retards reversal of learned model on the predation of artificial Batesian mimics. Anim Behav. avoidance. Oikos. 69:115–118. 25:660–665. Roper TJ, Redston S. 1987. Conspicuousness of distasteful prey affect Greenwood JJD, Cotton PA, Wilson DM. 1989. Frequency-dependent the strength and durability of one-trial avoidance learning. Anim selection on aposematic prey: some experiments. Biol J Linn Soc. Behav. 35:739–747. 36:213–226. Roper TJ, Wistow R. 1986. Aposematic colouration and avoidance Ham AD, Ihalainen E, Lindstro¨m L, Mappes J. 2006. Does colour learning in chicks. Q J Exp Psychol. 38B:141–149. matter? The importance of colour in avoidance learning, memora- Rowe C, Lindstro¨m L, Lyytinen A. 2004. The importance of pattern bility and generalisation. Behav Ecol Sociobiol. 60:482–491. similarity between Mu¨llerian mimics in predator avoidance learn- Huheey JE. 1988. Mathematical models of mimicry. Am Nat. 131:S22– ing. Proc R Soc Lond B. 271:407–413. S41. Rowland HM, Ihalainen E, Lindstro¨m L, Mappes J, Speed MP. 2007. Ihalainen E, Lindstro¨m L, Mappes J. 2007. Investigating Mu¨llerian Co-mimics have a mutualistic relationship despite unequal defen- mimicry: predator learning and variation in prey defences. J Evol ces. Nature. 448:62–67. Biol. 20:780–791. Ruxton GD, Sherratt TN, Speed MP. 2004. Avoiding attack—the evo- Ihalainen E. 2006. Experiments on defensive mimicry: linkages be- lutionary ecology of crypsis, warning signals and mimicry. 1st ed. tween predator behaviour and qualities of the prey [dissertation]. Oxford: Oxford University Press. Jyva¨skyla¨ (Finland): University of Jyva¨skyla¨. p. 37. Sherratt TN, Beatty CD. 2003. The evolution of warning signals as Jansson L, Enquist M. 2003. Receiver bias for colourful signals. Anim reliable indicators of prey defence. Am Nat. 162:377–389. Behav. 66:965–971. Sherratt TN, Speed MP, Ruxton GD. 2004. Natural selection on un- Kapan DD. 2001. Three-butterfly system provides a field test of mu¨l- palatable species imposed by state-dependent foraging behaviour. lerian mimicry. Nature. 409:338–340. J Theor Biol. 228:217–226. Kokko H, Mappes J, Lindstro¨m L. 2003. Alternative prey can change Shettleworth SJ. 1998. Cognition, evolution and behaviour. New York: model-mimic dynamics between parasitism and mutualism. Ecol Oxford University Press. Lett. 6:1068–1076. Sille´n-Tullberg B. 1985. Higher survival of an aposematic than of Langham GM. 2006. Rufous-tailed jacamars and aposematic butter- a cryptic form of a distasteful bug. Oecologia. 67:411–415. flies: do older birds attack novel prey? Behav Ecol. 17:285–290. Speed MP. 1993. Mu¨llerian mimicry and the psychology of predation. Lindstro¨m L, Alatalo RV, Lyytinen A, Mappes J. 2001a. Predator ex- Anim Behav. 45:571–580. perience on cryptic prey affects the survival of conspicuous apose- Speed MP. 1999. Batesian, quasi-Batesian or Mu¨llerian mimicry? The- matic prey. Proc R Soc Lond B. 268:357–361. ory and data in mimicry research. Evol Ecol. 13:755–776. Lindstro¨m L, Alatalo RV, Lyytinen A, Mappes J. 2001b. Strong anti- Symula R, Schulte R, Summers K. 2001. Molecular phylogenetic evi- apostatic selection against novel rare aposematic prey. Proc Natl dence for a mimetic radiation in Peruvian poison frogs supports Acad Sci. 98:9181–9184. aMu¨llerian mimicry hyphothesis. Proc R Soc Lond B. 268:2415– Lindstro¨m L, Lyytinen A, Mappes J, Ojala K. 2006. Relative impor- 2421. tance of taste and visual appearance for predator education in Turner JRG. 1977. Butterfly mimicry: the genetical evolution of an Mu¨llerian mimicry. Anim Behav. 72:323–333. adaptation. Evol Biol. 10:163–206. Mallet J, Joron M. 1999. Evolution of diversity in warning color and Wallace AR. 1867. Journal of Proceedings of the Entomological Soci- mimicry: polymorphisms, shifting balance, and speciation. Annu ety of London. In: Transactions of the Entomological Society of Rev Ecol Syst. 20:201–233. London. 3rd Series (years 1864–1869); lxxxviii–lxxxi.